Anthropic launches Claude Sonnet 4.5 — a coding- and agent-focused frontier model
Anthropic has released Claude Sonnet 4.5, a model positioned as an advance in coding, agentic tasks, reasoning, and math. The update pairs model improvements with several product changes aimed at longer-running, tool-enabled workflows and developer-focused integrations.
What’s new in products and tooling
Claude Sonnet 4.5 ships alongside a number of platform upgrades that target agentic and code-first use cases:
- Checkpoints in Claude Code to save progress and roll back to earlier states. (Details: https://anthropic.com/news/enabling-claude-code-to-work-more-autonomously)
- A refreshed terminal interface and a native VS Code extension (Marketplace: https://marketplace.visualstudio.com/items?itemName=anthropic.claude-code).
- Context editing and memory tools in the Claude API to support longer, more complex agent runs (https://anthropic.com/news/context-management).
- Code execution and file creation (spreadsheets, slides, documents) integrated into the Claude apps (download: https://claude.ai/redirect/website.v1.07f611e1-e39b-4e56-8251-b396f9288147/download).
- Expanded rollout of Claude for Chrome to Max users (https://www.anthropic.com/news/claude-for-chrome).
- The Claude Agent SDK exposes the same infrastructure that powers Claude Code for building custom agents (https://anthropic.com/engineering/building-agents-with-the-claude-agent-sdk).
Benchmarks and capabilities
Anthropic reports substantial gains on real-world coding and computer-use evaluations. Key figures called out include:
- SWE-bench Verified: 77.2% (reported using a two-tool scaffold, averaged over 10 trials with a 200K thinking budget on the 500-problem dataset).
- OSWorld: 61.4%, leading the benchmark and up from Sonnet 4’s 42.2% four months earlier.
- Observed ability to maintain sustained focus on complex, multi-step tasks for 30+ hours in practice.
The model also shows improvements across reasoning and math evaluations, and the company cites better domain-specific performance in finance, law, medicine, and STEM compared with earlier models.
Safety and alignment
Claude Sonnet 4.5 is released under Anthropic’s ASL-3 protections. The announcement highlights targeted safety work aimed at reducing concerning behaviors such as deception, sycophancy, and power-seeking, and improving defenses against prompt injection in agentic and computer-use contexts. Classifier-based filters focus on high-risk outputs (CBRN-related), and Anthropic notes reductions in false positives — described as a tenfold improvement since an earlier description and a twofold improvement since Opus 4.
Further technical details and alignment evaluations, including mechanistic interpretability tests, are available in the model’s system card (https://www.anthropic.com/claude-sonnet-4-5-system-card).
Developer access and pricing
Claude Sonnet 4.5 is stated to be available everywhere today. Developers can access the model as claude-sonnet-4-5 via the Claude API (https://docs.claude.com/en/docs/about-claude/models/overview). Pricing is reported to remain the same as Sonnet 4: $3 / $15 per million tokens, depending on rate tier.
The Claude Agent SDK and related Developer Platform updates are available for developers to build custom agents (https://claude.com/platform/api and https://anthropic.com/engineering/building-agents-with-the-claude-agent-sdk).
Research preview: Imagine with Claude
A limited research preview called “Imagine with Claude” demonstrates on-the-fly software generation and is available to Max subscribers for five days (https://claude.ai/redirect/website.v1.07f611e1-e39b-4e56-8251-b396f9288147/imagine).
Further reading
Full technical notes, evaluation methodology, and additional resources are posted on Anthropic’s site, including the model page and documentation:
- System card: https://www.anthropic.com/claude-sonnet-4-5-system-card
- Model page: https://www.anthropic.com/claude/sonnet
- API docs: https://docs.claude.com/en/docs/about-claude/models/overview
- Agent SDK engineering post: https://anthropic.com/engineering/building-agents-with-the-claude-agent-sdk
Original announcement: https://www.anthropic.com/news/claude-sonnet-4-5
